Yann LeCun, founder and chair of AMI (Advanced Machine Intelligence), expects artificial intelligence to tilt the balance in favor of cyberattackers only temporarily. Speaking on Tuesday at Semafor’s Silicon Valley & The World event, he said the world will probably move toward an era of unprecedented software-system safety.
LeCun told Semafor’s Shelly Banjo that AI tools could make cyberattacks easier and give attackers an advantage over defenders for a short time. He sees that as a transition, however: he expects AI eventually to strengthen software defenses as well.
World models and robots
The 2018 Turing Award winner has long been skeptical that simply scaling up large language models will lead to superintelligence or artificial general intelligence. LeCun left Meta in 2025 to establish AMI, a company focused on world models. It raised more than $1 billion in seed funding, a record for a European company. Referring to the size of the round, LeCun joked that something so large might better be called a coconut.
At the event, he also discussed applications of world models in physical AI, while acknowledging that researchers still have a long way to go in developing them. He said several conceptual breakthroughs will be needed before genuinely useful domestic robots can be built.
In LeCun’s view, humanoid robots can currently perform narrow tasks in carefully controlled environments, but household use is not imminent. He said he and other researchers working on world models are turning toward physical AI because it is a fundamental blind spot for large language models.
Open-source models
Asked about open-source models, LeCun also noted that the most powerful such systems are Chinese. He said this had disappointed many people in the United States, including some members of the current government.



